🤖 ReAct Data Analytics Agent

An intelligent data analytics agent powered by Google Gemini and the ReAct (Reasoning + Acting) framework

[

✨ Features

• 🧠 ReAct Reasoning: Systematic think-act-observe loops for robust analysis
• 💾 Dual Memory System: Short-term conversation + long-term dataset storage
• 🛠️ 5 Powerful Tools: Load, inspect, analyze, and visualize data
• 🔄 Auto Retry Logic: Handles rate limits with exponential backoff
• 📊 LLM-Generated Code: Gemini creates pandas/matplotlib code on the fly
• 🎨 Interactive UI: Chat-like interface in Google Colab
• 🔒 Safe Execution: Sandboxed code runner prevents system access

🚀 Quick Start

Option 1: Google Colab (Recommended)

1. Click the "Open in Colab" badge above
2. Add your Gemini API key
3. Run all cells
4. Start analyzing data!

Option 2: Local Installation

git clone https://github.com/YOUR_USERNAME/react-data-agent.git
cd react-data-agent
pip install -r requirements.txt

💡 Usage Examples

Load and Analyze Data

from src.agent import DataAnalyticsAgent
import google.generativeai as genai

# Initialize
genai.configure(api_key="YOUR_API_KEY")
model = genai.GenerativeModel("gemini-1.5-pro")
agent = DataAnalyticsAgent(model)

# Analyze
result = agent.run("Load insurance.csv and show me average premium by region")
print(result["answer"])

Interactive Interface

from src.interface import setup_datasets_with_ui, ColabAgentInterface

# Load data with UI
setup_datasets_with_ui()

# Start chat interface
interface = ColabAgentInterface(agent)
interface.display()

🏗️ Architecture

User Query ↓ ┌─────────────────────┐ │ DataAnalyticsAgent │ │ (ReAct Loop) │ └─────────────────────┘ ↓ ┌─────────────────────┐ │ Reasoning (LLM) │ ← Google Gemini │ • Think │ │ • Plan │ │ • Decide │ └─────────────────────┘ ↓ ┌─────────────────────┐ │ Tool Selection │ │ • load_csv │ │ • analyze │ │ • visualize │ │ • inspect │ └─────────────────────┘ ↓ ┌─────────────────────┐ │ Code Executor │ ← Safe sandbox │ • Run pandas │ │ • Generate plots │ └─────────────────────┘ ↓ Result → Memory → Next Iteration

📊 Example Queries

• "Load sales.csv and show me total revenue by product"
• "What are the top 5 customers by purchase frequency?"
• "Create a bar chart of monthly sales trends"
• "Which regions have the highest average order value?"
• "Show me correlation between age and insurance premium"

🧠 How It Works

The agent uses the ReAct (Reasoning + Acting) pattern:

1. THINK: "User wants sales trends. I need to check if data is loaded."
2. ACT: Execute list_datasets() tool
3. OBSERVE: "No datasets loaded yet"
4. THINK: "I need to load the data first"
5. ACT: Execute load_csv(filepath="sales.csv")
6. OBSERVE: "Data loaded successfully with 1000 rows"
7. THINK: "Now I can analyze trends"
8. ACT: Execute analyze(query="calculate monthly sales")
9. OBSERVE: "Analysis complete: [results]"
10. THINK: "I have the answer"
11. ACT: DONE with final answer

🛠️ Components

Core Agent (src/agent.py)

• Orchestrates ReAct loop
• Manages iterations and memory
• Handles retry logic

Tools (src/tools/)

• LoadCSVTool: Load datasets into memory
• ListDatasetsTool: View available data
• InspectDatasetTool: Examine structure
• AnalyzeTool: Generate pandas code
• VisualizeTool: Create charts

Memory (src/memory.py)

• Short-term: Recent conversation (50 turns)
• Long-term: Loaded datasets (persistent)

Executor (src/executor.py)

• Safe code execution
• Sandboxed environment
• Prevents file system access

🔧 Configuration

agent = DataAnalyticsAgent(
    gemini_model=model,
    max_iterations=20,      # Max reasoning steps
    verbose=True,           # Print reasoning
    retry_delay=2.0,        # Rate limit backoff
    max_retries=3           # Max retry attempts
)

📈 Performance

• Average query time: 10-30 seconds
• Typical iterations: 3-8 steps
• Rate limit handling: Automatic retry with backoff
• Memory usage: ~50MB per 1M row dataset

🐛 Troubleshooting

Rate Limit (429) Errors

The agent automatically handles rate limits with exponential backoff. If you still encounter issues:

• Increase retry_delay (default: 2.0s)
• Increase max_retries (default: 3)
• Wait 60 seconds between complex queries

"Dataset not found" Errors

Always load data first:

agent.run("Load mydata.csv as 'data'")
agent.run("Now analyze the data")  # Works!

Code Generation Issues

If the LLM generates incorrect code:

• Use inspect_dataset to show column names
• Be specific in your query
• Try rephrasing the question

📄 License

This project is licensed under the MIT License - see LICENSE file.

🙏 Acknowledgments

• ReAct Paper by Yao et al.
• Google Gemini API
• Anthropic for Claude (inspiration)